AI Factory Glossary

recency bias, training phenomena

Recent examples influence more.

recurrent memory transformer, llm architecture

Combine attention with external memory.

recurrent memory transformer,llm architecture

Combine recurrence with attention.

recurrent state space models, rssm, reinforcement learning

Model dynamics with recurrent and stochastic components.

recurrent video models, video understanding

RNN/LSTM for temporal modeling.

recursive forecasting, time series models

Recursive forecasting uses one-step-ahead model iteratively feeding predictions as inputs for multi-step forecasts.

recursive reward modeling, ai safety

Supervise reward model with another model.

recursive reward, ai safety

Recursive reward modeling trains reward models using other reward models hierarchically.

red teaming, ai safety

Adversarial testing by human experts.

red teaming,ai safety

Adversarial testing to find model vulnerabilities weaknesses or harmful behaviors.

red-teaming, ai safety

Red-teaming probes models for harmful behaviors informing safety training.

reference image conditioning, generative models

Use reference image to guide.

reference image, multimodal ai

Reference images guide generation by transferring style or content through adapter networks.

referring expression comprehension, multimodal ai

Find objects from descriptions.

referring expression generation, multimodal ai

Generate descriptions of objects.

reflection agent, ai agents

Reflection enables agents to critique their own plans and outputs identifying improvements.

reflexion,ai agent

Agent learns from feedback and mistakes by generating reflections and improving.

reformer, llm architecture

Reformer uses locality-sensitive hashing for approximate attention matching.

reformer,foundation model

Use LSH attention to reduce complexity from quadratic to linear.

refusal behavior, ai safety

Model declining to answer.

refusal calibration, ai safety

Balance safety and helpfulness.

refusal training, ai safety

Refusal training explicitly teaches models when to decline requests.

refusal training, ai safety

Teach model when to refuse.

refused bequest, code ai

Subclass not using inherited methods.

regenerative thermal, environmental & sustainability

Regenerative thermal oxidizers recover heat through ceramic beds improving energy efficiency.

regex constraint, llm optimization

Regex constraints enforce pattern matching during text generation.

region-based captioning, multimodal ai

Caption specific image areas.

regret minimization,machine learning

Goal in online learning to minimize cumulative mistakes.

reinforcement graph gen, graph neural networks

Reinforcement learning guides graph generation through rewards for desired properties like drug-likeness.

reinforcement learning for nas, neural architecture

Use RL to search architectures.

reject option,ai safety

Allow model to say "I don't know".

relation extraction as pre-training, nlp

Use relation extraction as pretext task.

relation networks, neural architecture

Networks explicitly modeling pairwise relations.

relation-aware aggregation, graph neural networks

Relation-aware aggregation weights messages by edge types when updating node representations.

relational knowledge distillation, rkd, model compression

Transfer relationships between samples.

relative position bias, transformer

Learn relative position embeddings.

reliability-centered maintenance, rcm, production

Optimize maintenance strategy.

relm (regular expression language modeling),relm,regular expression language modeling,structured generation

Constrain generation using regex patterns.

renewable energy credits, environmental & sustainability

Renewable energy credits represent environmental attributes of renewable electricity generation.

renewable energy, environmental & sustainability

Renewable energy sources like solar and wind are increasingly adopted by fabs to reduce carbon footprint and energy costs.

renewal process, time series models

Renewal processes model event sequences where inter-arrival times are independent identically distributed random variables.

rényi differential privacy, training techniques

Rényi differential privacy generalizes epsilon-delta framework using Rényi divergence.

reorder point, supply chain & logistics

Reorder point is the inventory level triggering replenishment orders calculated from lead time demand and desired service level.

repetition penalty, llm optimization

Repetition penalty reduces probability of recently generated tokens.

replaced token detection, rtd, foundation model

Training objective in ELECTRA.

replanning, ai agents

Replanning revises strategies when plans fail or conditions change.

replica exchange, chemistry ai

Parallel simulations at different temperatures.

replicate,model hosting,simple

Replicate hosts ML models with simple API. Community models. Easy to deploy custom models.

repository understanding,code ai

Analyze entire codebases to understand architecture and dependencies.

representational similarity analysis, rsa, explainable ai

Compare representations across models.